Over the last decade, Unmanned Aerial Vehicles (UAVs) have been extensively used in many commercial applications due to their manageability and risk avoidance. One of the main problems considered is the Mission Planning for multiple UAVs, where a solution plan must be found satisfying the different constraints of the problem. This problem has multiple variables that must be optimized simultaneously, such as the makespan, the cost of the mission or the risk. Therefore, the problem has a lot of possible optimal solutions, and the operator must select the final solution to be executed among them. In order to reduce the workload of the operator in this decision process, a Decision Support System (DSS) becomes necessary. In this work, a DSS consisting of ranking and filtering systems, which order and reduce the optimal solutions, has been designed. With regard to the ranking system, a wide range of Multi-Criteria Decision Making (MCDM) methods, including some fuzzy MCDM, are compared on a multi-UAV mission planning scenario, in order to study which method could fit better in a multi-UAV decision support system. Expert operators have evaluated the solutions returned, and the results show, on the one hand, that fuzzy methods generally achieve better average scores, and on the other, that all of the tested methods perform better when the preferences of the operators are biased towards a specific variable, and worse when their preferences are balanced. For the filtering system, a similarity function based on the proximity of the solutions has been designed, and on top of that, a threshold is tuned empirically to decide how to filter solutions without losing much of the hypervolume of the space of solutions.

Low-Rank Adaptation (LoRA), a parameter-efficient fine-tuning method that leverages low-rank adaptation of weight matrices, has emerged as a prevalent technique for fine-tuning pre-trained models such as large language models and diffusion models. Despite its huge success in practice, the theoretical underpinnings of LoRA have largely remained unexplored. This paper takes the first step to bridge this gap by theoretically analyzing the expressive power of LoRA. We also quantify the approximation error when the LoRArank is lower than the threshold. All our theoretical insights are validated by numerical experiments. Recent foundation models, such as large language models (OpenAI, 2023; Liu et al., 2019; Touvron et al., 2023), have achieved remarkable success in a wide range of applications. Due to their substantial size, the standard full fine-tuning approach--where all the model's parameters are updated for specialized tasks--is becoming increasingly difficult and inefficient. This leads to the growing popularity of parameter-efficient fine-tuning approaches (Hu et al., 2022a; Liu et al., 2022; Ben Zaken et al., 2022; Hu et al., 2022b). Instead of updating all parameters, these approaches selectively update smaller subsets of weights or introduce lightweight adapters, thereby greatly decreasing the computational and storage costs. The most dominant approach along this line is Low-Rank Adaptation (LoRA) (Hu et al., 2022a), which employs lightweight low-rank adapters to pre-trained weight matrices. Far from merely enhancing computational efficiency, empirical evidence has shown that LoRA can match or even exceed the performance of full fine-tuning (Hu et al., 2022a). To date, LoRA has been widely used and achieved considerable success in adapting large language models (Hu et al., 2022a; Dinh et al., 2022b) and image generation models (Ryu, 2023; Fan et al., 2023) for various downstream tasks. Despite the empirical success of LoRA, little is known in theory about how it works. In fact, several crucial theoretical questions remain open, such as: What is the minimum rank of the LoRA adapters required to adapt a (pre-trained) model f to match the functionality of the target model f? How does the model architecture (i.e., depth, width) affect the minimal rank? If the adapter rank is lower than this threshold, what is the resulting approximation error?

Convolutional Neural Networks (CNNs) are widely used in fault diagnosis of mechanical systems due to their powerful feature extraction and classification capabilities. However, the CNN is a typical black-box model, and the mechanism of CNN's decision-making are not clear, which limits its application in high-reliability-required fault diagnosis scenarios. To tackle this issue, we propose a novel interpretable neural network termed as Time-Frequency Network (TFN), where the physically meaningful time-frequency transform (TFT) method is embedded into the traditional convolutional layer as an adaptive preprocessing layer. This preprocessing layer named as time-frequency convolutional (TFconv) layer, is constrained by a well-designed kernel function to extract fault-related time-frequency information. It not only improves the diagnostic performance but also reveals the logical foundation of the CNN prediction in the frequency domain. Different TFT methods correspond to different kernel functions of the TFconv layer. In this study, four typical TFT methods are considered to formulate the TFNs and their effectiveness and interpretability are proved through three mechanical fault diagnosis experiments. Experimental results also show that the proposed TFconv layer can be easily generalized to other CNNs with different depths. The code of TFN is available on https://github.com/ChenQian0618/TFN.

The amount of data on the internet is doubling every two years and 90% of it is video, the internet's infrastructure has reached a point of saturation. While there is massive data traffic, there is very little bandwidth to handle it, and the compression technologies that lie at its heart are buckling under pressure. London-based AI startup Deep Render has built a revolutionary video compression algorithm that shrinks video sizes up to 5x without compromising the streaming quality. Instead of improving on the traditional video compression systems, Deep Render has completely reinvented the technology to mimic the neural processes of the human eye. In a recent development, Deep Render, which operates with the mission to solve the data and bandwidth crisis threatening the future of the web, has raised $9 million in funding for its transformational video compression technology.

Seldon, a London-based data-centric machine learning operations (MLOps) platform, has secured a $20M Series B funding round led by new Portuguese investor Bright Pixel (former Sonae IM) with participation from existing investors AlbionVC (backed Ophelos), Cambridge Innovation Capital, and Amadeus Capital Partners. The funding will help Seldon expand its machine learning product's market fit and unlock enterprise-ready solutions based on open source. "AI is in everything, and Seldon is uniquely positioned to ensure a return on ML investment by providing robust, scalable, and secure infrastructure, pioneering a data-centric approach to ML pipelines, prioritizing team collaboration across the organization, and making sure teams can solve meaningful problems at scale by building trust in machine learning, even under the most intense regulatory conditions. "We're excited to bring together new investor Bright Pixel Capital and our existing partners, who believe in our vision and can help us become the trusted MLOps partner of any organization worldwide." Currently, numerous companies are investing a lot of resources into artificial intelligence, but they are having difficulty expanding their models for practical use. This is due to bottlenecks in team workflows, increased regulation and compliance restraints, a lack of trust in model outputs, and ensuring peak model performance are all top of mind for AI-powered enterprises. Here's where Seldon helps Data Scientists, ML Engineers, and other stakeholders in the company to quickly and efficiently adopt machine learning to address these challenges. Founded in 2014, Seldon is a data science and machine learning operations platform that aims to empower Data Scientists, ML Engineers, and MLOps teams to deploy, monitor, explain, and manage their ML models. With Seldon, organisations can minimise risk and drastically cut down time-to-value from their models. The UK company offers both an open-source framework, "Core," which focuses on model deployment, and an enterprise product, "Deploy Advanced," which builds on this functionality to power model monitoring, explainability, and management. Seldon claims that it has achieved a 400% YoY growth rate in its open-source frameworks installed and running since its series A in November 2020. "Seldon has differentiated itself by presenting a unique solution that can reduce the friction for users deploying and explaining ML models across any industry.

Cordless, a London-based modern telephony software with conversation intelligence, has raised $2.3 million in a seed funding round led by Berlin-based VC Fly Ventures with participation from Passion Capital (also backed Atoa Payments), TrueSight (also backed BondAval, Claimer), and Renaud Visage, co-founder of Eventbrite (also backed Mason and Climate X). The UK startup will use the funds to accelerate customer acquisition, and product development and deepen integrations in the customer support ecosystem. Luba Chudnovets and Irina Bednova, who met while building Monzo bank working as a Head of Scaling Operations and a Technical Lead, respectively, founded Cordless. In Monzo, the female founders worked on the problem of scaling customer support when Monzo was going through exponential growth. During their stint in Monzo, the female duo learned an important point -- if you build products that customers want to use, it's essential to have a deep understanding of "how your customers feel and why they reach out for support."

A precision health company developing medical imaging tools to improve the diagnosis of metabolic diseases and cancer, Perspectum, has completed the first close of its Series C funding round. The company's $36 million investment round was led by Oppenheimer Holdings. With this, the total funding raised by the company accounts for $120 million. Perspectum is working to expand its footprint across the country and grow its customer base. In addition to scaling up its US operations, Perspectum will use the new funding to accelerate its product pipeline for multiorgan inflammatory conditions and oncology.

Volta Medical, a France-based health technology company developing AI solutions to assist electrophysiologist physicians and surgeons, has secured €36 million in Series B funding. With this, the total funding raised by the company accounts for €70 million. The investment round was led by the US-based Vensana Capital alongside participation from Lightstone Ventures (which backed Dunzo and Nimbus Therapeutics) and existing investor Gilde Healthcare. The funding will help accelerate new product development, support additional clinical trials, prepare for full-scale US commercialisation, and pursue further regulatory approvals. The company's lead product, VOLTA VX1, is the first commercially available AI decision-support software to help guide physicians with identification and real-time annotation of unique abnormalities on 3D anatomical and electrical maps of the heart.

ChAI, a London-based startup that uses artificial intelligence (AI) and machine learning to predict commodity prices, has secured a seed round to help expand its industry-leading services into new markets, including raw-material businesses and key supply chain providers. The funding will enable them to invest in developing commodity risk insurance products that will enable smaller companies that are underserved by existing risk transfer solutions, such as options or futures markets, to transfer commodity price risks and derivatives. It will also enable larger companies and material providers to transfer financial risk for a variety of raw materials, including plastic and steel. ChAI employs unique data sets, such as satellite imagery and maritime transport data, to provide unprecedented visibility and confidence in price fluctuations and movements of the world's key raw materials, such as oil, aluminium, copper, and others. Founded by Marcus Dixon, Michael Button, Silvi Wompa, Stephen Butler, Tristan Fletcher in 2019, ChAI is democratising tools, techniques, and data while combining cutting-edge AI techniques with new alternative data sources to eliminate risk in physical supply chains.

The future of defence aviation is autonomous. Recently Sheild AI joined the race and now Razor's Edge, based in Reston, Virginia, and defence- and security-focused VC firm announced the closing of its third startup investment fund at under $340M. The firm investment indicates that national security technology is a safe bet even in difficult economic times. According to the firm, it has surpassed its initial target of $250M and will target companies developing autonomous systems, space technologies, cybersecurity, AI and machine learning, digital signal processing, and other aerospace and defence technologies. With the new funding, the firm's total assets under management now exceed $600M.